Beruflich Dokumente
Kultur Dokumente
Javed Bari
Jong-Suk Jung
Yow-Shi Ho
Date: 12/04/00
Foreword
As part of the course IEE572, this group was looking forward to designing an
engineering experiment. Luckily, the members of this group were already engaged in an
experiment in the Advanced Asphalt Laboratory of the Department of Civil and
Environmental Engineering of Arizona State University. So the group members decided to
design that experiment. In that time, only the first quarter of the course content of the IEE572
course was covered. With that background knowledge, it was not possible to make a
comprehensive design plan. So the group decided to design a factor screening experiment.
As each test took a considerably long time, it was not possible to perform complementary
detailed experiments during the time period of this semester. All the three group-members
actively participated in the design and experiment works.
Introduction
Background of the Experiment
In pavement engineering, permanent deformation (p) is an important parameter
regarding design. Many researchers have earlier observed that test result of p varies with the
size and geometry of test specimens of uniaxial load test using universal testing machine. For
composite materials like asphalt concrete, the ideal test specimen must be large enough
relative to the size of the individual aggregate particles. But the problem is that different
researchers have been using test specimens of different size and geometry for test purpose.
This leads to different sets of database often not compatible to each other. So it becomes hard
to compare one result to another and even harder to use them as basis for further analysis.
These inconsistencies are basically due to the absence of an available and universally agreed
optimum specimen size and geometry. In fact few researches has so far been conducted for
obtaining an optimum specimen size and geometry. As part of their graduate research, the
group members had already been involved in a similar research, though in an early stage. So
the group decided to design and conduct a factor screening experiment with a view to finding
out an optimum specimen size and geometry for testing permanent deformation of asphalt
pavement.
The Experiment
Objectives
The objective of the experiment is to determine the important and optimum factors,
along with their correlation, for a test specimen to be tested in permanent deformation test of
asphalt pavements, which should lead to fairly accurate and consistent permanent deformation
values.
Test Method
The experiment is a uniaxial load test that was conducted with universal testing
machine in the in the Advanced Asphalt Laboratory of the Department of Civil and
Environmental Engineering of Arizona State University. The test specimens were fabricated
in the Superpave Gyratory Compactor. The loading type is repetitive and uniaxial. The
preparation of samples, testing of specimens and collection of data were consistent with the
standard procedures.
Experimental Plan
Hypothesis
The experimental hypotheses are:
1) Minimum specimen dimension has a significant effect on permanent deformation
measured in the uniaxial test.
2) There is a limiting minimum specimen dimension above which the material properties are
independent of this dimension. This dimension is the optimum dimension.
These hypotheses address the concept of homogeneity of the test specimens.
Replication
The test was replicated four times. As each test takes a considerably long time, it was
not feasible to plan for replication more than four.
Analysis procedures
The analysis procedure was determined in accordance with the standard procedure for
a 2K factorial design outlined in the text Design and Analysis of Experiments written by
Douglas C. Montgomery. The planned analysis steps are outlined in the next page.
Experimental Design
The final experimental design is shown below in Table 1.
Table 1 Summary of the Experimental Design
Standard
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
Run
sequence
27
10
7
19
25
28
16
4
18
12
1
17
31
30
13
8
26
11
2
14
6
23
20
21
3
22
24
15
32
29
9
5
12.5
12.5
12.5
12.5
37.5
37.5
37.5
37.5
12.5
12.5
12.5
12.5
37.5
37.5
37.5
37.5
12.5
12.5
12.5
12.5
37.5
37.5
37.5
37.5
12.5
12.5
12.5
12.5
37.5
37.5
37.5
37.5
70
70
70
70
70
70
70
70
150
150
150
150
150
150
150
150
70
70
70
70
70
70
70
70
150
150
150
150
150
150
150
150
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
PD
Test Result
The test results are summarized in Table 2 below.
Table 2 Summary Test Results
Std
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
Run
27
10
7
19
25
28
16
4
18
12
1
17
31
30
13
8
26
11
2
14
6
23
20
21
3
22
24
15
32
29
9
5
A
12.5
12.5
12.5
12.5
37.5
37.5
37.5
37.5
12.5
12.5
12.5
12.5
37.5
37.5
37.5
37.5
12.5
12.5
12.5
12.5
37.5
37.5
37.5
37.5
12.5
12.5
12.5
12.5
37.5
37.5
37.5
37.5
B
70
70
70
70
70
70
70
70
150
150
150
150
150
150
150
150
70
70
70
70
70
70
70
70
150
150
150
150
150
150
150
150
C
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
PD
6850
6592
6218
7491
13427
9781
11780
10669
5334
5541
5673
5757
5888
4413
5524
8358
6234
6076
5663
6820
7109
10330
8826
8469
4760
5033
5101
4789
4190
6806
5498
5992
Analysis
Initial analysis
This experiment is a 23 design with 4 replicates. That means there are three factors (A,
B and C) each with two levels; one low and the other is high (as shown in the experimental
plan). The three factors that are of our interest are A(Nominal aggregate size), B(Specimen
diameter) and Hight to diameter (H/D). Table 3 shows the analysis of varience (ANOVA) that
summarizes the effect estimates and sum of squares. It is evident from the ANOVA that A, B,
C, and AB interaction are significant at about the 5 percent level.
1047.22
6905.88
15.16
4.679E+007
Prob > F
< 0.0001 significant
< 0.0001
significant
< 0.0001
significant
0.0066
significant
0.0005
significant
0.2092
0.1436
0.1224
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision 12.401
0.8299
0.7803
0.6976
Model Refinement
As the half normal plot reveals that only A, B, C and AB interaction are significant, a
reanalysis was performed. Table 4 shows the revised ANOVA after removing the
nonsignificant factors.
H a lf N o r m a l p lo t
99
B
Half Normal % probability
97
95
A
AB
90
85
80
70
60
40
20
0
0 .0 0
6 8 2 .5 3
1 3 6 5 .0 6
2 0 4 7 .5 9
2 7 3 0 .1 3
|E ffe c t|
Adeq Precision
13.408
10
F
Value
24.46
27.66
48.11
7.81
14.25
Prob > F
< 0.0001
< 0.0001
< 0.0001
0.0094
0.0008
2.17
0.1176
0.7837
0.7517
0.6962
significant
not significant
Factor
Intercept
A-A(NAS)
1.00
B-B(SD)
C-C(H/D)
AB
Coefficient
Estimate
6905.88
1035.13
DF
1
1
-1365.06
-550.13
-742.81
1
1
1
Standard
95% CI
Error
Low
196.81
6502.06
196.81 631.31
196.81-1768.88
196.81 -953.94
196.81-1146.63
95% CI
High
7309.69
-961.24
-146.31
-338.99
VIF
1.00
1.00
1.00
The revised predictive equation in terms of actual factors are shown below:
PD = 6154.45313 + 246.22875 * A(NAS) +3.01406 * B(SD) - 1100.25 * C(H/D)
- 1.48562 * A(NAS) * B(SD)
Diagonostic Checking
Figure 2 is the normal probability plot of the residuals. There is clearly no ploblem
with normality. Figure 3 is the plot of the residuals versus the predicted values. Figure 4 is
the plot of residuals versus run number. It may be noticed that there are problems in model
fitting in Figure 3 and Figure 4. A data transformation is often used to deal with such
plroblems. So several data transformation were tried.
11
N o r m a l p lo t o f r e s id u a ls
Normal % probability
99
95
90
80
70
50
30
20
10
5
1
-2 . 3 4
-1 . 0 6
0 .2 1
1 .4 9
2 .7 7
S t u d e n t i z e d R e si d u a l s
Studentized Residuals
3 .0 0
1 .5 0
0 .0 0
-1 . 5 0
-3 . 0 0
4 6 9 8 .3 8
6DESIGN-EX
1 7 3 .5 3
7 6 PERT
4 8 . 6 9 Plot
9 1 2 3 .8 4
1 0 5 9 9 .0 0
P re d i c t e d
PD
12
R e s id u a ls v s . R u n
Studentized Residuals
3 .0 0
1 .5 0
0 .0 0
-1 . 5 0
-3 . 0 0
11
16
21
26
31
Final ANOVA
The final ANOVA is shown in Table 5. Figure 6 is the normal probability plot of the
residuals. Figure 7 is the plot of the residuals versus the predicted values. Figure 8 is the plot
of residuals versus run number. There is no ploblem with these plots. Figure 9 is the plot of
predicted versus actual values of response variable. It shows good plot around the line of
equality, indeed. Figure 10 and Figure 11 represent one factor plot and a contour plot
respectively.
DESIGN-EXPERT Plot
13
B o x- C o x P lo t fo r P o w e r T r a n s fo r m s
1 8 .8 7
Ln(ResidualSS)
1 8 .4 0
1 7 .9 3
1 7 .4 6
1 6 .9 9
-3
-2
-1
Lam bda
14
DESIGN-EXPERT Plot
Constant:
F
Value
22.71
21.13
54.55
6.62
8.53
0.48
0.7709
0.7369
0.6781
Prob > F
< 0.0001 significant
< 0.0001
< 0.0001
0.0159
0.0070
0.6979
not significant
PRESS
2.928E-005
Adeq Precision
13.019
Factor
Intercept
A-A(NAS)
B-B(SD)
C-C(H/D)
AB
Coefficient
Estimate
0.012
-7.139E-004
1.147E-003
3.997E-004
4.536E-004
DF
1
1
1
1
1
Standard
95% CI
95% CI
Error
Low
High
1.553E-004
0.012
0.013
1.553E-004 -1.033E-003 -3.952E-004
1.553E-004 8.285E-004 1.466E-003
1.553E-004 8.105E-005 7.184E-004
1.553E-004 1.349E-004 7.723E-004
=
*A
*B
*C
*A*B
15
VIF
1.00
1.00
1.00
1.00
N o r m a l p lo t o f r e s id u a ls
99
95
Normal % probability
90
80
70
50
30
20
10
5
1
-2 .4 4
-1 .1 7
0 .1 1
1 .3 8
2 .6 6
S t u d e n ti z e d R e si d u a l s
Studentized Residuals
1 .5 0
0 .0 0
-1 .5 0
-3 .0 0
0 .0 1 0
0 .0 1 1
0 .0 1 2
0 .0 1 3
0 .0 1 4
P re d i c te d
DESIGN-EXPERT Plot
16
R e s id u a ls v s . R u n
3 .0 0
Studentized Residuals
1 .5 0
0 .0 0
-1 . 5 0
-3 . 0 0
11
16
21
26
31
P r e d ic te d v s . A c tu a l
0 .0 1 5
Predicted
0 .0 1 4
0 .0 1 2
0 .0 1 0
0 .0 0 9
0 .0 0 9
0 .0 1 0
0 .0 1 2
0 .0 1 4
0 .0 1 5
A c tu a l
17
O n e F a c to r P lo t
13427
PD
1 1 1 1 7 .8
8 8 0 8 .5
6 4 9 9 .2 5
4190
1 .0 0
1 .2 5
1 .5 0
1 .7 5
2 .0 0
C : C (H / D )
1 5 0 .0 0
PD
B: B(SD)
1 3 0 .0 0
5 9 9 0 .0 6
1 1 0 .0 0
6 7 5 0 .4
9 0 .0 0
7 5 1 0 .7 4
8 2 7 1 .0 7
7 0 .0 0
1 2 .5 0
4
1 8 .7 5
2 5 .0 0
3 1 .2 5
3 7 .5 0
A : A (N A S )
18
Conclusion
From the analysis it has been found that the higher level of H/D ratio gives better result. The
contour plot reveals that low level of A and high level of B lead to a optimum result. The final
predictive equation is as follows:
1/Sqrt(PD) = 0.011989 -1.56901E-004 * A(NAS) + 6.00119E-006 * B(SD)
+7.99475E-004 * C(H/D) +9.07147E-007 * A(NAS) * B(SD)
19